Available models

Qwen3.7 Flash

-15%

Alibaba Cloud

A fast multimodal Qwen3.7 model with stronger vision, agent execution, and coding throughput than Qwen3.6 Flash. Built for low-latency multimodal understanding and agent workflows.

Qwen3.7 Max

-15%

Alibaba Cloud

The Qwen3.7 flagship, optimized for coding, productivity, and agent-centric workloads. Its agent-oriented design emphasizes coding, office productivity, and complex multi-step work.

Qwen3.7 Plus

-15%

Alibaba Cloud

A cost-efficient multimodal Qwen model for general reasoning, vision, and production applications. It combines text and image understanding with a cost-conscious design for production-scale multimodal reasoning.

Qwen3.8 Flash

-10%

Alibaba Cloud

A fast multimodal Qwen3.8 model with a native 1M context for coding, agents, documents, and visual understanding. It approaches flagship quality at a fraction of Qwen3.8 Max cost, with image and video input.

Qwen3.8 Max

Alibaba Cloud

A 2.4T-parameter MoE flagship for coding, office productivity, vision, and long-horizon agents. Native multimodal understanding supports long documents and video, with strong end-to-end delivery on professional tasks.

Claude Fable 5

Flex -50%

Anthropic

An Anthropic model built for autonomous knowledge work, coding, and multimodal agents. It is positioned for autonomous knowledge work and coding with multimodal input, reasoning, and tool-oriented workflows.

Claude Fable 5.1

Flex -50%

Anthropic

Anthropic's successor to Claude Fable 5 for long-running agentic coding, multistep research, and document work. Adaptive thinking stays on for sustained coding, research, and document workflows.

Claude Haiku 4.5

Anthropic

A fast, efficient Claude model for latency-sensitive chat, coding, and agent tasks. It targets near-frontier intelligence at substantially lower latency and cost for high-volume interactive workloads.

Claude Opus 4.6

Flex -50%

Anthropic

A powerful Claude model for end-to-end coding and long-running professional workflows. It is optimized for agents that operate across complete software and professional workflows instead of isolated prompts.

Claude Opus 4.7

Flex -50%

Anthropic

An Opus model designed for long-running asynchronous agents and complex software workflows. The model is designed for asynchronous agents that sustain coding and professional work over extended runs.

Claude Opus 4.8

Flex -50%

Anthropic

A high-capability Opus model for long-context reasoning, coding, and multimodal analysis. It combines multimodal input, reasoning support, and a long context window for demanding general-purpose workflows.

Claude Opus 5

Flex -50%

Anthropic

Anthropic's flagship for demanding reasoning, coding, visual analysis, and long-running agents. It is especially capable at end-to-end software tasks, code review and bug finding, document and chart analysis, complex office deliverables, tool use, and coordinating parallel subagents.

Claude Opus 5.5

Flex -50%

Anthropic

Anthropic's model for long-running agentic coding and knowledge work. Adaptive thinking stays on, with a default medium effort level for sustained coding and professional tasks.

Claude Sonnet 4.6

Flex -50%

Anthropic

A balanced frontier model for iterative development, agents, and professional knowledge work. Strengths include iterative development, complex codebase navigation, and end-to-end project execution with tools.

Claude Sonnet 5

Flex -50%

Anthropic

A frontier Sonnet model for coding, professional work, and adaptive agentic reasoning. Adaptive reasoning levels let applications balance frontier coding and agent performance against latency and token use.

Doubao Seed 2.1 Pro

ByteDance

A newer Doubao Pro model for complex reasoning, coding, and multimodal production use.

DeepSeek V4 Pro

Peak & off-peak-20%

DeepSeek

The official DeepSeek V4 Pro flagship for advanced reasoning, coding, and sustained agent workloads. The 0813 production checkpoint keeps 1M context and dual thinking modes under the same API name.

DeepSeek V4.1 Flash

Peak & off-peak-20%

DeepSeek

The official DeepSeek V4.1 Flash model for fast long-context reasoning, coding, and agents. Evaluations show it surpassing V4 Pro on performance, cost, speed, and total time.

Gemini 3 Flash Preview

Google

A fast Gemini thinking model for multi-turn chat, coding, and agent workflows. It offers near-Pro reasoning and tool use at Flash-class speed for multi-turn chat, coding, and agents.

Gemini 3.1 Flash Lite

Google

A high-efficiency Gemini preview optimized for high-volume, latency-sensitive applications. It is tuned for high-volume workloads while approaching the quality of larger Flash models.

Gemini 3.1 Pro Preview

Flex -70%

Google

A frontier Gemini reasoning preview for software engineering and reliable agentic workflows. The preview improves software-engineering performance, agent reliability, and token efficiency across complex workflows.

Gemini 3.5 Flash Lite

Google

A lightweight Gemini model for focused subagents and high-volume production tasks. The lightweight design is especially suited to focused subagents inside larger multi-agent systems.

Gemini 3.6 Flash

Google

A high-efficiency Gemini model for coding, agents, and web or application development. It focuses on polished coding and application outputs with fewer unnecessary edits across agentic workflows.

Gemini 3.8 Flash

Default -50%Flex -70%

Google

Google's most capable Flash model for long-horizon software engineering, autonomous agents, and complex enterprise workflows. It improves software engineering, agent tasks, and multi-step reasoning over Gemini 3.7 Flash.

Kimi K2.6

Moonshot AI

A multimodal Kimi model for long-horizon coding, UI generation, and agent orchestration. It targets long-horizon coding, code-driven UI generation, and coordinated multi-agent execution across complex projects.

Kimi K2.7 Code

Moonshot AI

A coding-focused Kimi model for reliable end-to-end programming across long contexts. The coding-focused design aims to complete end-to-end programming tasks reliably across long contexts.

Kimi K3

Moonshot AI

An open-weight multimodal Kimi model for complex coding, reasoning, and agentic work. The open-weight multimodal reasoning model is positioned for complex coding, knowledge work, and long-running agents.

GPT-5.4

OpenAI

A flagship GPT model for advanced reasoning, coding, multimodal input, and professional workflows. It unifies general GPT and coding capabilities for large-context professional work and tool-driven execution.

GPT-5.4 mini

OpenAI

A faster GPT-5.4 variant for high-throughput reasoning, coding, and multimodal applications. The smaller variant keeps strong reasoning and coding capabilities while improving speed and throughput.

GPT-5.4 nano

OpenAI

A compact GPT model optimized for low-latency, high-volume, and cost-sensitive tasks. It is tuned for speed-critical classification, extraction, routing, and other high-volume lightweight tasks.

GPT-5.5

OpenAI

A frontier GPT model for complex professional work with stronger reasoning and reliability. It strengthens reliability and token efficiency on difficult professional tasks while retaining long-context support.

GPT-5.6 Luna

Flex -70%

OpenAI

A fast, efficient GPT-5.6 model for chat, classification, and lightweight agents. The efficient tier targets high-volume chat, classification, and lightweight agent workflows with low latency.

GPT-5.6 Sol

Default -20%Flex -70%

OpenAI

The GPT-5.6 flagship for demanding reasoning, coding, and multi-step agent tasks. The flagship tier is particularly strong at command-line work and complex multi-step coding workflows.

GPT-5.6 Terra

Flex -70%

OpenAI

A balanced GPT-5.6 model for everyday coding, reasoning, and agentic applications. The balanced tier sits between flagship quality and cost efficiency for everyday coding, reasoning, and agents.

GPT-6 Astra

Flex -70%

OpenAI

OpenAI's most capable model for demanding end-to-end work across reasoning, coding, computer use, research, and document creation. The flagship tier is built for complex professional workflows that need the highest reasoning effort.

GPT-6 Luna

Flex -70%

OpenAI

OpenAI's most efficient GPT-6 model for focused, high-volume tasks. The efficient tier targets high-concurrency chat, classification, and lightweight agent workflows with low latency.

GPT-6 Sol

Flex -70%

OpenAI

A GPT-6 model for complex coding and agentic workflows. It balances flagship capability with lower cost, and supports adjustable reasoning effort for multi-step coding and tool-driven agents.

Grok 4.5

Flex -50%

xAI (Grok)

A general-purpose Grok model for coding, knowledge work, STEM, and demanding reasoning. It supports both reasoning and non-reasoning modes for agentic software and workflow tasks.

Grok 4.6

Flex -50%

xAI (Grok)

A frontier Grok model for long-running agents, coding, and knowledge work. It focuses on multi-step agents, ambitious interactive and visual work, and sustained coding trajectories.

GLM-5.2

-20%

Zhipu AI

A long-context GLM reasoning model for project-level software engineering and agents. The large-scale reasoning model is intended for project-level software engineering and long-horizon agent workflows.

GLM-5.3

Zhipu AI

A flagship GLM model for long-horizon coding, agents, and security review. It keeps the GLM-5.2 base and raises coding and agent performance through post-training.

Doubao Seedream 5.0

-20%

ByteDance

A Seedream model for general image generation, editing, and visual design workflows.

Doubao Seedream 5.0 Pro

-20%

ByteDance

A premium Seedream model for high-fidelity image generation and professional visual creation.

GPT Image 2

OpenAI

OpenAI's high-fidelity image generation and editing model for the dedicated Images API.

GPT Image 2.5 Flare

OpenAI

OpenAI's fastest model for high-quality everyday image generation and editing. It accepts text and image inputs and is tuned for responsive, high-volume visual workflows.

GPT Image 2.5 Sunburst

OpenAI

OpenAI's image model for workflows where editing precision matters most. It generates and edits images from text and image inputs when fine control over the result is the priority.